Skip to content

OpenEdit

A local video editor built for AI agents: the agent makes the creative calls, OpenEdit handles frames, previews and rendering.

Reviewed by Sheela · Updated

Setup docs Open source

Our rating

3.4/5

A promising design that lets the agent edit footage rather than describe edits, from a project too young to rely on for client work.

What it is

OpenEdit is a desktop video editor for Windows and Linux that an AI agent drives through MCP. You select a moment and describe the edit; the agent inspects timestamped frames and transcript ranges instead of reading whole videos, then applies edits, graphics, transitions and overlays. OpenEdit owns frame timing, media processing, previews and export, using FFmpeg and MLT.

The MCP surface is six tools: query_project, edit_project, run_script for atomic Python edits, trigger_render, get_render_job and cancel_render_job. Edits are reversible, final exports never overwrite earlier ones, and local Whisper transcription is optional. It is MIT-licensed; v1.3.1 shipped on 11 August 2026, and the project is about three months old with 52 stars.

Install

Release ZIP plus install script; open_edit mcp-config --project <path>

Our review

Most AI video tools either generate clips or tell you which buttons to press. OpenEdit gives an agent a real timeline. The split of work is sensible: the agent decides what to cut and caption, and the editor does the frame-accurate parts models are bad at.

The details show care. Agents look at bounded, timestamped frames and transcript ranges rather than repeatedly ingesting a whole video, which keeps context use sane. run_script applies a batch of edits atomically with a timeout, so a half-applied change does not leave a broken project. The installer runs a real edit, transition, overlay, preview and export before reporting success.

The limits are those of a three-month-old, single-maintainer project with 52 stars. There is no macOS build. GPU encoding depends on the driver, and the CPU renderer is the guaranteed path. The README does not claim it understands every request, and nothing independent has tested it yet.

As an idea it is ahead of most of the field: the agent never needs a model key inside the editor, because your agent host handles its own provider authentication. That makes it easy to try with Claude Code, Codex or OpenCode on a short recording before trusting it with anything longer.

Where it earns its place

  • A developer advocate turning a 20-minute demo recording into a two-minute clip

    Cut dead air, add a title card and export

    Connect Claude Code via the generated mcp.json, ask for the cuts against the transcript, review the preview and trigger the final render. Free apart from model usage.

  • A small YouTube channel on Linux

    Rough-cut interviews from the transcript

    Enable the optional Whisper install, have the agent mark sections to keep from transcript ranges, then polish by hand in the editor.

Before you commit

Setup
Download the Windows x64 or Linux x64 release, run install.ps1 or install.sh; it provisions Python, Node, FFmpeg/MLT and Chromium, then tests a full edit. Python 3.12 recommended.
Integrations
Writes MCP config for generic clients, OpenCode and Codex; six MCP tools; optional built-in chat via OpenCode.
Data and privacy
Runs locally; projects live in ~/OpenEditProjects. Optional Whisper transcription downloads model data. MIT.
Limits
No macOS build; GPU encoder support varies by driver; three months old with one open issue and few users.

What's good

  • Agent edits a real timeline, not a description
  • Atomic scripted edits and reversible history
  • Installer verifies a full edit-to-export run
  • No model key needed inside the editor

What's not

  • Windows and Linux only
  • Young project with a single maintainer
  • GPU encoding not guaranteed across drivers

Best for

Developers and creators who want to experiment with agent-driven editing of short recordings.

Alternatives

OpenEdit: common questions

Is OpenEdit worth it?
A promising design that lets the agent edit footage rather than describe edits, from a project too young to rely on for client work. We rate it 3.4 out of 5.
How much does OpenEdit cost?
Free and open source under MIT; your agent host pays for its own model usage.
What are the downsides of OpenEdit?
Windows and Linux only; young project with a single maintainer; gPU encoding not guaranteed across drivers.
Who is OpenEdit best for?
Developers and creators who want to experiment with agent-driven editing of short recordings.
Is OpenEdit safe for company data?
Runs locally; projects live in ~/OpenEditProjects. Optional Whisper transcription downloads model data. MIT.
What are the best OpenEdit alternatives?
The closest alternatives we have reviewed are MCP for Blender.

Answers checked Oct 24, 2026.